EasyTPP is a comprehensive toolkit for Temporal Point Processes (TPPs), designed for researchers and practitioners to facilitate development, benchmarking, and application of TPP models. It offers a unified framework for various TPP architectures, supporting both TensorFlow and PyTorch, and aims to improve reproducibility and ease of customization.

How It Works

EasyTPP employs a modular design with abstract classes, allowing for easy integration and development of new TPP models. It provides implementations for numerous state-of-the-art TPP papers, including RMTPP, NHP, and THP, with compatibility across TensorFlow (v1.13.1 and v2.0+) and PyTorch (v1.7.0+). This dual framework support caters to both academic research and industrial applications.

Quick Start & Requirements

• Installation: pip install easy-tpp or clone from GitHub and run python setup.py install.
• Prerequisites: Python 3.7.0+, TensorFlow or PyTorch. Datasets need to be preprocessed and placed in a specific directory structure.
• Resources: Requires dataset preparation. Documentation generation requires Sphinx.
• Links: Colab Tutorials, OnlineDoc - Datasets, OnlineDoc - QuickStart.

Highlighted Details

• Supports 8 state-of-the-art TPP models with TensorFlow and PyTorch implementations.
• Includes 5 real-world and 2 non-anthropogenic datasets, plus HuggingFace dataset API integration.
• Provides an Optuna-based hyper-parameter optimization pipeline.
• Paper accepted at ICLR 2024.

Maintenance & Community

The project is jointly initiated by Machine Intelligence Group, Alipay, and DAMO Academy, Alibaba. Notable related projects include EasyRec, EasyNLP, and FuxiCTR.

Licensing & Compatibility

Licensed under the Apache License (Version 2.0). Some included code may be under different open-source licenses; refer to the NOTICE file.

Limitations & Caveats

The project is actively developed, with a "Todo List" indicating planned additions like Meta Temporal Point Process and Model-based RL via TPP.